1. Soft Walls: Algorithms to Enforce Aviation Security Adam Cataldo Prof. Edward Lee Prof. Shankar Sastry NASA JUP January 22-23, 2004 NASA Ames, Mountain View, CA Center for Hybrid and Embedded Software Systems

2. Outline The Soft Walls system Objections Control system design Current Research Conclusions

3. A Deadly Weapon? Project started September 11, 2001

4. Introduction On-board database with “no-fly-zones” Enforce no-fly zones using on-board avionics

5. Dragonfly 3 Dragonfly 2 Ground Station Early Prototype Using Stanford DragonFly UAVs [Claire Tomlin, Jung Soon Jang, Rodney Teo]

6. Flight Test Result Nov 19 th, 2003 Moffett Federal Air Field

7. Another Early Prototype, Demo’d by Honeywell on National TV, Dec., 2003 Based on advanced ground avoidance system Issues a warning when approaching terrain or a no-fly zone Takes over control from the pilot when approach is too close Returns control to the pilot after diverting Demonstrated on ABC World News Tonight Dec. 30, 2003 Honeywell pilot on ABC World News Tonight with Peter Jennings, Dec. 30, 2003.

8. Both Prototypes use Autonomous Control Pilot or Path Planning Controller Aircraft Soft Walls controller

9. Our End Objective is Not Autonomous Control but a Blending Controller Pilot Aircraft Soft Walls + bias pilot control as needed

10. Our End Objective Maximize Pilot Authority, but keep the aircraft out of forbidden airspace

11. Unsaturated Control No-fly zone Control applied Pilot remains neutral Pilot turns away from no-fly zone Pilot tries to fly into no-fly zone

12. In the News ABC World News Tonight with Peter Jennings –Dec. 30, 2003 Radio Interviews –Voice of America, Dec. 6, 2003. –NPR Marketplace –WTOP, Washington DC, July 14, 2003 –As It Happens, CBC, July 9, 2003 Magazines –New Scientist, July 2, 2003 –Salon, December 13, 2001 –Slashdot, July 3, 2003 –Slashdot, Jan 3, 2004 Newspapers –New York Times, April 11, 2002 –Toronto Globe and Mail –The Washington Times –The Orlando Sentinel –The Straits Times (Singapore) –The Times of India –The Star (South Africa) –The Age (Australia) –Reuters, July 2, 2003 Graphic on ABC World News Tonight with Peter Jennings, Dec. 30, 2003.

14. Objections Reducing pilot control is dangerous –reduces ability to respond to emergencies

15. There is No Emergency That Justifies Attempting to Land on Fifth Ave. Although there are clearly regions of space where flying is absolutely unacceptable, regulatory restraint is required to avoid overconstraining the air space. Today, some pilot responses to emergencies can result in a passenger aircraft being shot down.

16. Objections Reducing pilot control is dangerous –reduces ability to respond to emergencies There is no override –pilots want a switch in the cockpit

17. There are already regions of space for which no override switch enables transit Terrain imposes “hard wall” constraints on airspace. We are proposing that spaces be defined that are as surely constrained but more gently enforced. Again, regulatory restraint is required to not overconstrain the airspace.

18. Objections Reducing pilot control is dangerous –reduces ability to respond to emergencies There is no override –pilots want a switch in the cockpit Localization technology can fail –GPS can be jammed

19. Localization Backup Radio beacons Inertial navigation –drift limits accuracy –affects the geometry of no-fly zones

20. Objections Reducing pilot control is dangerous –reduces ability to respond to emergencies There is no override –pilots want a switch in the cockpit Localization technology can fail –GPS can be jammed Deployment could be costly –Software certification? Retrofit older aircraft?

21. Deployment Fly-by-wire aircraft –a software change –which is of course extremely costly Older aircraft –autopilot level? –Honeywell prototype? Phase in –prioritize airports

22. Objections Reducing pilot control is dangerous –reduces ability to respond to emergencies There is no override –pilots want a switch in the cockpit Localization technology could fail –GPS can be jammed Deployment could be costly –how to retrofit older aircraft? Complexity –software certification

23. Not As Complex as Air Traffic Control Self-contained avionics system (not multi- vehicle) Human factors is an issue: –pilot training? –air traffic controller training?

24. Objections Reducing pilot control is dangerous –reduces ability to respond to emergencies There is no override –pilots want a switch in the cockpit Localization technology could fail –GPS can be jammed Deployment could be costly –how to retrofit older aircraft? Deployment could take too long –software certification Fully automatic flight control is possible –throw a switch on the ground, take over plane

25. Potential Problems with Ground Control Human-in-the-loop delay on the ground –authorization for takeover –delay recognizing the threat Security problem on the ground –hijacking from the ground? –takeover of entire fleet at once? Requires radio communication –hackable –jammable

26. Relationship to Flight Envelope Protection With flight envelope protection, the limits on pilot-induced maneuvers are known Knowing these limits enables tighter tolerances, and hence tighter geometries for no-fly zones. see http://softwalls.eecs.berkeley.edu for FAQ

27. Here’s How It Works

28. Previous Algorithm: What We Want to Compute No-fly zone Backwards reachable set States that can reach the no-fly zone even with Soft Walls controller Can prevent aircraft from entering no-fly zone

29. Theorem [Computing ]: where is the unique viscosity solution to: The Backwards Reachable Set for the Stanford no-fly Ellipse

30. What We Create The terminal payoff function l:X -> Reals The further from the no-fly zone, the higher the terminal payoff payoff northward position eastward position no-fly zone (constant over heading angle) - +

31. What We Compute No-fly zone terminal payoff Backwards Reachable Set optimal payoff

32. Our Control Input Backwards Reachable Set optimal control at boundary dampen optimal control away from boundary State Space optimal payoff function

33. How we computing the optimal payoff (analytically) We solve this equation for J*: Reals^n x [ 0, ∞) -> Reals J* is the viscosity solution of this equation J* converges pointwise to the optimal payoff as T-> ∞ (Tomlin, Lygeros, Pappas, Sastry) dynamicsspatial gradient time derivative terminal payoff

34. (Mitchell) Computationally intensive: n states  O(2^n) How we computing the optimal payoff (numerically) northward position eastward position heading angle no-fly zone time 01M

35. Discretize Time Control Algorithm—At Each Step 1.Compute safe control inputs for next N steps 2.Calculate optimal control input for next N steps 3.Use only the first optimal input Current Research: Model Predictive Control control input pilot input and noise ukuk u k+1 …u k+N

36. Computing Safe Control Inputs (Pappas) Given: Compute: We assume: control inputspilot inputs no-fly zone safe control inputs If the next N control inputs are in the safe set, then the state will remain outside the no-fly zone.

37. Calculating Optimal Input We want the control input to equal zero whenever possible We want the control input to change slowly from each input to the next We minimize, over the safe control inputs,

38. Dragonfly 3Dragonfly 2 Ground Station Stanford DragonFly UAVs [Claire Tomlin, Jung Soon Jang, Rodney Teo]

39. Another Experimental Platform In collaboration with the Penn UAV team

40. Conclusions Embedded control system challenge Control theory identified Future design challenges identified http://softwalls.eecs.berkeley.edu

41. Acknowledgements Ian Mitchell George Pappas Xiaojun Liu Shankar Sastry Steve Neuendorffer Claire Tomlin